Apeireth: An Agentic OS in Rust with Topological Memory and Its Own Kernel

At the end of September 2026, the Apeireth repository appeared on GitHub — an agent environment that the authors describe as an "AGI Operating System & Cognitive Microkernel" written in pure safe Rust. In 28 days, the project gained 303 stars and three contributors. This is not a library for a single agent, but rather an operating system for cognitive processes: the kernel owns the canonical turn protocol and abstract ports, while specific tools, adapters, and SQLite work are moved into separate crates.

What it does

• Its own tools and functions — the package includes a ProcessExecutor with resource limits via JobObject/cgroups and a RepoMap for building an AST map of the repository, which allows the agent to launch processes and analyze code.
• Agent memory — hybrid memory search combines BM25, dense vector cosine distances, and RRF fusion across 10,000 nodes; the median P50 latency is claimed at 3.65 ms.
• Human approval — there is a separate endpoint for authorizing L0-level actions, meaning human-in-the-loop is built into the protocol.
• Voice and speech — the perception module includes Whisper speech recognition and MiniMax TTS synthesis, although the documentation marks it as being in the process of being connected.
• Images and documents — the same perception module claims screen vision, that is, analysis of images from the screen.

Its own model instead of the cloud

The platform is written in Rust and connects its own model via an OpenAI-compatible API: the documentation specifies the endpoint POST /v1/chat/completions with buffered SSE chat streaming. This means any local server that can expose such an interface will work. License — Apache License 2.0: you can use, modify, distribute, and sublicense it, including for commercial purposes, provided you retain the authorship and warranty disclaimer notices.

How it differs from its neighbors

Of the three neighbors in the registry, only Apeireth claims to handle images and documents via screen vision. At the same time, MCP support, multiple agents working together, and its own server with Docker — features that most neighbors have — are not mentioned in Apeireth's documentation. In terms of how it connects its own model, Apeireth is closer to DeerFlow (also an OpenAI-compatible API), whereas goose is limited to Ollama, and Hermes Agent offers the widest set: Ollama, llama.cpp, and vLLM.

How alive the project is

The repository was created on 09/04/2026, meaning the project is less than a month old. In that time, two stable releases have come out, the latest being 1.5.0 on 09/04/2026. There are three contributors and 303 stars. For comparison: goose has 54 thousand stars and 14 releases in 90 days, Hermes Agent has 247 thousand stars and 16 releases, and DeerFlow has 83 thousand stars and one release over the same period.

Apeireth will suit those looking for an agent environment in Rust with a focus on long-term memory, a cognitive kernel, and the ability to connect their own model via an OpenAI-compatible API. It is convenient to follow new agent platforms in the CPU3D news feed.